Artificial intelligence

Corporate AI · behavior policy and prompt library

Set the language, the tone and the guidelines artificial intelligence should follow across the organization, and keep a library of approved prompts. The policy is appended to the instructions sent to the model before every AI call.

Without an agreed rule, every AI feature answers its own way: one in English, another too casual, another naming a customer in a message meant to go outside. The corporate AI policy standardizes the behavior in one place, and it applies to every AI feature in the product at once.

Open Settings and click Corporate AI.

Corporate policy

Turn on the Enable corporate policy switch. While it is off, and it starts off, nothing about the AI's behavior changes.
Under Response language, pick the language the AI should always use, or leave it on Free (follows the input).
Under Tone, write the tone you want, for example professional, warm.
Under Guidelines (guardrails), write in free text what the AI should and should not do.
Adjust the Allow customer data and Allow financial data switches.
Save.

From then on, every AI call the product makes carries those instructions along, ahead of the request itself. That applies to the assistant, to the suggestions, to the AI agents and to the other features that generate text.

Example: A consultancy serving international clients turns the policy on, sets the language to English, the tone to professional, and writes in the guideline that the AI must never promise deadlines or discounts. From then on, a rep who asks for an email draft in Portuguese gets the text in English, in the agreed tone, and with no promise about timing.

An important caveat about the two data switches

Allow customer data and Allow financial data are instructions written for the model, not technical blocks. Turning them off adds to the request an order not to cite or infer customers' personal data, or not to cite amounts and revenue, and to answer in aggregate or qualitative terms instead. The record's content is still sent to the AI provider exactly as before.

If your goal is to stop a piece of data from leaving your organization, this switch is not the path. Use field-level permission to hide the data from whoever triggers the AI, and the removal of personal data before the request reaches the model, which is a separate AI security setting.

Prompt library

In the Prompt library card, fill in the Title and, if you want, the Category.
Write the text under Prompt content.
Click Add prompt.

The library is your organization's own catalog, meant to standardize the requests that work and keep every person from reinventing the same text. Be clear about its reach: the prompts stored here are consulted on this screen, for the team to copy and reuse. They are not injected automatically into AI calls and they do not show up as an option inside the assistant.

Who can do this

Saving the policy and creating, editing or deleting library prompts requires permission to manage settings.
Reading the policy and the library is available to anyone who opens the screen.
Every save of the policy is written to the audit trail as Configured AI, along with the state of the switches.

💡 Guardrails work best when they are short, specific and verifiable. A sentence like do not promise delivery dates or discounts without approval is followed far more consistently than a generic paragraph about quality.

Common questions

Does turning the policy off erase what I wrote? No. The text stays stored and applies again when you turn the switch back on.
Does the policy apply to calls made through the API? It is applied to the AI calls the product makes.
Where do I control how much AI each person spends? Under Settings, AI usage.

Open this article inside the system →